Cartoon Sketch Colorization

Abstract

Manual sketch colorization remains a time consuming and labor intensive bottleneck in animation and digital art production. Traditional workflows demand extensive human effort and offer limited creative scalability, posing significant challenges for artists and studios particularly in resource constrained environments like Pakistan’s growing animation sector. This project addresses these inefficiencies through an intelligent, AI powered web platform designed for automated sketch colorization and text to image generation. The proposed system integrates a custom trained Pix2Pix U-Net model for sketch colorization and a Stable Diffusion pipeline for text guided image generation. The methodology follows an Agile development approach, including data pre processing, model training and refinement, multi-modal integration, and performance evaluation using metrics such as SSIM, PSNR, MAE, and MSE. Users can either upload sketches for instant colorization or input descriptive text to generate original visuals through a seamless, chat style interface. Evaluation results indicate a 25.1% reduction in MAE and a 12.4% improvement in SSIM compared to baseline models, demonstrating enhanced visual quality and model accuracy. Designed with modular architecture and responsive design, the platform empowers digital artists, educators, and content creators by accelerating ideation, enhancing productivity, and democratizing access to generative AI. This project contributes a novel, accessible solution with transformative potential for the global creative technology landscape